About

Patrick Charbonneau is a professor of chemistry and physics at Duke University. Originally from Montreal, he earned his Ph.D. in chemical physics from Harvard University in 2006 under the supervision of David Reichman. Following his doctoral studies, he was a Marie-Curie Fellow with Daan Frenkel at Amolf in Amsterdam before moving to Durham in 2008. Charbonneau's primary research interests lie in the fields of soft matter and statistical physics. He employs theory and numerical simulations to study colloidal and protein self-assembly and has been a principal investigator of the Simons Collaboration on Cracking the Glass Problem. To date, he has authored over 100 peer-reviewed publications and has co-organized several international meetings on these topics. In addition to his scientific research, Charbonneau is a science history enthusiast. He curated the exhibit "Seeing the Invisible: 50 Years of Macromolecular Visualization," leads an oral history project titled "The History of Replica Symmetry Breaking in Physics," and is co-editing a book about Women in the History of Quantum Physics. Occasionally, he teaches courses on the science of cooking and the history of chemistry. His interest in the history of North American confections reflects the interplay of these various pursuits. Since joining Duke University, Charbonneau has received several honors, including a National Science Foundation CAREER Award, a Sloan Fellowship, and the Oak Ridge National Lab Ralph E. Powe award. He has held various visiting scientist positions and has twice been named a Top Reviewer by the Journal of Chemical Physics. He is also a fellow of the American Physical Society (APS).

Research topics

• Condensed matter physics
• Physics
• Computer Science
• Quantum mechanics
• Statistical physics
• Theoretical physics
• Thermodynamics
• Law
• Epistemology
• Mechanics

Selected publications

• The critical slowing down in diffusion models

  arXiv (Cornell University) · 2026-05-12

  preprintOpen access

  Computational sampling has been central to the sciences since the mid-20th century. While machine-learning-based approaches have recently enabled major advances, their behavior remains poorly understood, with limited theoretical control over when and why they succeed. Here we provide such insight for diffusion models-a class of generative schemes highly effective in practice-by analyzing their application to the $O(n)$ model of statistical field theory in the Gaussian limit $n \to \infty$. In this analytically tractable setting, we show that training a score model with a one-layer network architecture matching the exact solution exhibits a form of critical slowing down in parameter learning. This slowing down also impacts the generation process, indicating that the well-known difficulties of sampling near criticality persist even for learned generative models. To overcome this bottleneck, we demonstrate the power of combining architectural depth with physical locality. We find that using a two-layer architecture drastically reduces the critical slowing down, with the training time scaling logarithmically rather than quadratically with system size. By introducing a local score approximation we show that this acceleration in training time can be achieved without increasing the number of neural network parameters. Taken together, these results demonstrate that diffusion models can overcome the critical slowing down through appropriate architectural design, and establish a controlled framework for understanding and improving learned sampling methods in statistical physics and beyond.

  PublisherDOI

• The critical slowing down in diffusion models

  ArXiv.org · 2026-05-12

  articleOpen access

  Computational sampling has been central to the sciences since the mid-20th century. While machine-learning-based approaches have recently enabled major advances, their behavior remains poorly understood, with limited theoretical control over when and why they succeed. Here we provide such insight for diffusion models-a class of generative schemes highly effective in practice-by analyzing their application to the $O(n)$ model of statistical field theory in the Gaussian limit $n \to \infty$. In this analytically tractable setting, we show that training a score model with a one-layer network architecture matching the exact solution exhibits a form of critical slowing down in parameter learning. This slowing down also impacts the generation process, indicating that the well-known difficulties of sampling near criticality persist even for learned generative models. To overcome this bottleneck, we demonstrate the power of combining architectural depth with physical locality. We find that using a two-layer architecture drastically reduces the critical slowing down, with the training time scaling logarithmically rather than quadratically with system size. By introducing a local score approximation we show that this acceleration in training time can be achieved without increasing the number of neural network parameters. Taken together, these results demonstrate that diffusion models can overcome the critical slowing down through appropriate architectural design, and establish a controlled framework for understanding and improving learned sampling methods in statistical physics and beyond.

  PublisherOA PDF

• Soil environmental DNA versus pitfall trapping for small mammal monitoring: qPCR assay development and pilot field evaluation

  Canadian Journal of Zoology · 2026-01-01

  article

  Small mammals are fundamental drivers of ecosystem processes, and they act as bioindicators of ecological integrity. Although their monitoring is crucial from a conservation perspective, conventional surveying devices such as pitfall traps are invasive and raise ethical concerns. A potential non-invasive alternative is the analysis of environmental DNA (eDNA) shed by small mammals in the environment. The first step in the adoption of eDNA-based monitoring is the design of species-specific, sensitive, and efficient quantitative PCR (qPCR) assays. Here, we developed and validated targeted qPCR assays for three Canadian small mammals: the smoky shrew ( Sorex fumeus Miller, 1895), the masked shrew ( Sorex cinereus Kerr, 1792), and the southern bog lemming ( Synaptomys cooperi Baird, 1858). We also conducted a pilot field study to compare small mammal detection rates from pitfall trap capture and eDNA analysis of soil samples. For the smoky shrew and the masked shrew, detection rates were low for both pitfall traps (6% and 16%, respectively) and eDNA (6% and 3%, respectively), while no capture or eDNA detection was made for the southern bog lemming, indicating that a greater sampling effort would be required to better compare the two methods. Additional studies are thus needed before eDNA can be adopted as a reliable method for monitoring small mammals.

  PublisherDOI

• Editorial: <i>Physical Review E</i>—A Uniquely Diverse Journal

  Physical review. E · 2025-02-11

  editorialOpen access1st authorCorresponding
  PublisherOA PDFDOI

• Elizabeth Monroe Boggs: From Quantum Chemistry to the Manhattan Project

  Cambridge University Press eBooks · 2025-06-19

  book-chapter1st authorCorresponding
  PublisherDOI

• Amorphous Packings of Spheres

  2025-08-27 · 1 citations

  preprintOpen accessSenior author

  The apparent simplicity of amorphous sphere packings can be misleading. Although random close packing is a common phenomenon that has been studied for decades, recent theoretical and computational advances reveal how complex the underlying physics actually is. By bringing together perspectives from thermodynamics, constraint satisfaction problems, and optimization, unique insights emerge. This chapter reviews these advances and describes a framework for addressing some of the remaining open problems, including how to define a more rigorous analogue to random close packing and how to compute its properties.

  PublisherOA PDFDOI

• Fusion or Confusion? Assessing the impact of visible-thermal image fusion for automated wildlife detection

  ArXiv.org · 2025-11-27

  preprintOpen access

  Efficient wildlife monitoring methods are necessary for biodiversity conservation and management. The combination of remote sensing, aerial imagery and deep learning offer promising opportunities to renew or improve existing survey methods. The complementary use of visible (VIS) and thermal infrared (TIR) imagery can add information compared to a single-source image and improve results in an automated detection context. However, the alignment and fusion process can be challenging, especially since visible and thermal images usually have different fields of view (FOV) and spatial resolutions. This research presents a case study on the great blue heron (Ardea herodias) to evaluate the performances of synchronous aerial VIS and TIR imagery to automatically detect individuals and nests using a YOLO11n model. Two VIS-TIR fusion methods were tested and compared: an early fusion approach and a late fusion approach, to determine if the addition of the TIR image gives any added value compared to a VIS-only model. VIS and TIR images were automatically aligned using a deep learning model. A principal component analysis fusion method was applied to VIS-TIR image pairs to form the early fusion dataset. A classification and regression tree was used to process the late fusion dataset, based on the detection from the VIS-only and TIR-only trained models. Across all classes, both late and early fusion improved the F1 score compared to the VIS-only model. For the main class, occupied nest, the late fusion improved the F1 score from 90.2 (VIS-only) to 93.0%. This model was also able to identify false positives from both sources with 90% recall. Although fusion methods seem to give better results, this approach comes with a limiting TIR FOV and alignment constraints that eliminate data. Using an aircraft-mounted very high-resolution visible sensor could be an interesting option for operationalizing surveys.

  PublisherOA PDFDOI

• Not-So-Glass-Like Caging and Fluctuations of an Active Matter Model

  Physical Review Letters · 2025-06-04 · 3 citations

  preprintOpen accessSenior author

  Simple active models of matter recapitulate complex biological phenomena. The out-of-equilibrium nature of these models, however, often makes them beyond the reach of first-principle descriptions. This limitation is particularly perplexing when attempting to distinguish between different glass-forming mechanisms. We here consider a minimal active system in various spatial dimensions to identify the processes underlying their sluggish dynamics. Activity is found to markedly impact cage escape processes and critical fluctuations associated with exploring lower-dimensional caging features.

  PublisherOA PDFDOI

• Percolation and criticality of systems with competing interactions on Bethe lattices: limitations and potential strengths of cluster schemes

  ArXiv.org · 2025-10-03

  preprintOpen access

  The random clusters introduced by Fortuin and Kasteleyn (FK) and analyzed by Coniglio and Klein (CK) for Ising and related models have led first Swendsen and Wang and then Wolff to formulate remarkably efficient Markov chain Monte Carlo sampling schemes that weaken the critical slowing down. In frustrated models, however, no standard way to produce a comparable gain at small frustration -- let alone efficiently sample the large frustration regime -- has yet been identified. In order to understand why formulating appropriate cluster criteria for frustrated models has thus far been elusive, we here study minimal short-range attractive and long-range repulsive as well as spin-glass models on Bethe lattices. Using a generalization of the CK approach and the cavity-field method, the appropriateness and limitations of the FK--CK type clusters are identified. We find that a standard, constructive cluster scheme is then inoperable, and that the frustration range over which generalized FK--CK clusters are even definable is finite. These results demonstrate the futility of seeking constructive cluster schemes for frustrated systems but leaves open the possibility that alternate approaches could be devised.

  PublisherOA PDFDOI

• The geometry of jamming algorithms in the random Lorentz gas

  Proceedings of the National Academy of Sciences · 2025-11-05 · 2 citations

  articleOpen accessCorresponding

  Deterministic optimization algorithms unequivocally partition a complex energy landscape into inherent structures (ISs) and their respective basins of attraction. Can these basins be defined solely through geometric principles? This question is paramount to understanding hard sphere jamming, a key model of disordered matter. We here address the issue by proposing a geometric class of gradient descent-like algorithms, which we use to study a system in the hard-sphere universality class, the random Lorentz gas. The statistics of the resulting ISs is found to be strictly inherited from those of Poisson-Voronoi tessellations. The landscape roughness is further found to give rise to a hierarchical organization of ISs, which various algorithms explore differently. In particular, greedy and reluctant schemes tend to favor ISs of markedly different densities. The resulting ISs nevertheless robustly exhibit a universal force distribution, thus confirming the geometric nature of the jamming universality class. Along the way, the physical origin of a dynamical Gardner transition is identified.

  PublisherDOI

Recent grants

• Soft Matter Simulation and Theory of the Crystal Assembly of Globular and Membrane Proteins

  NSF · $270k · 2018–2022

• CAREER: Soft Matter Self-Assembly: Protein Crystallization and Colloidal Microphase Formation

  NSF · $450k · 2011–2016

Frequent coauthors

• Francesco Zamponi

  Sapienza University of Rome

  99 shared

• Giorgio Parisi

  53 shared

• Ludovic Berthier

  Gulliver

  44 shared

• Yi Hu

  35 shared

• Giulio Biroli

  École Normale Supérieure - PSL

  34 shared

• Yuliang Jin

  Institute of Theoretical Physics

  26 shared

• Gilles Tarjus

  Centre National de la Recherche Scientifique

  25 shared

• Lin Fu

  HKUST Shenzhen Research Institute

  22 shared

Labs

• Charbonneau LabPI

  Soft matter and statistical physics, colloidal and protein self-assembly

Awards & honors

• Journal of Chemical Physics Top Reviewer (2016, 2018)
• Top 20 Reviewers for 2012 (2013)
• Alfred P. Sloan Research Fellow (2013)
• Mention of Teaching Excellence, Duke University (2012)
• Open Eye Award, American Chemical Society (2011)